Research On Regional Short Term Wind Power Prediction Based On Feature Selection And Data Mining

Wind power has the characteristics of randomness and volatility.Due to the large-scale wind power integration,wind power brings great disturbance to the power system,which is a big challenge for the safe and stable operation of the power grid.Regional wind power prediction is a powerful measure to improve the wind energy consumption.From the perspective of feature selection and data mining in the paper,the regional short-term wind power prediction method is improved.The main contents include four parts: regional wind feature construction,regional wind feature selection,regional short-term wind power prediction based on feature selection,and regional short-term wind power prediction based on multifeature similarity matching method.The regional wind feature construction based on various mathematical methods is carried out.Through the source analysis and data preprocessing of the original data,the effectiveness of the data is improved.Through the feature construction of the preprocessed wind power data,the high-dimensional feature database of regional wind power is constructed,and the information of the original wind data is fully mined.A multi-stage feature selection model of regional wind power is proposed.Based on three feature selection methods,a multi-stage feature selection model of regional wind power is established in the paper.The appropriate feature subsets are separately selected in different stages.The feature subsets are evaluated with the power prediction model.The optimal feature subset of regional wind power prediction is finally obtained.The regional short-term wind power prediction model based on xgboost is established.Based on the results of feature selection,the key parameters of xgboost are optimized and the factors restricting the short-term prediction accuracy in the existing model are analyzed.Then the influence of the model generalization ability and the number of input features on the shortterm prediction effect are studied.Finally,a short-term regional wind power prediction model with strong generalization ability is obtained and the short-term prediction accuracy is improved based on the model.A wind power cluster power prediction model based on multi-feature similarity matching is proposed.The features distance calculation formula of the original model is improved based on the introduced power regulation coefficient.In the paper,the four key parameters of the model are optimized and the prediction results based on multi-feature similarity matching method are compared with that based on the traditional method,which shows that the prediction accuracy of the proposed model is significantly improved.